Exposing the Dangers of Distance: Mining Crash Narratives to Explore Why Pedestrians Face Severe Injury and Death Far From Home
Pedestrian fatalities in the United States have risen by 83% over the past 15 years, with much of the increase occurring on multilane suburban arterials. In Tennessee, deaths nearly tripled between 2009 and 2022, with studies linking crashes to high-speed midblock locations lacking pedestrian infrastructure. Spatial analysis shows pedestrians are being struck farther from home: in 2014 the median distance between residence and crash site was 1.5 miles, compared to four miles by 2023, while the share of crashes within one mile of home fell from 46% to 30%. Relative to city centers, crash locations remain stable, but the distance between city centers and pedestrian residences has grown, indicating that more crashes involve individuals living farther from urban cores. These shifts suggest pedestrians are traveling into distant, high-risk environments, raising essential questions about why they are walking in such areas and what broader urban trends contribute to this exposure. This study applies a hybrid methodology combining structured crash records with insights from unstructured police narratives from Tennessee’s Integrated Traffic Analysis Network (2014–2024). A home-based approach links pedestrian and driver addresses with U.S. Census block group characteristics, including income levels, vehicle ownership, education, commuting modes, and housing density, to better understand who is involved in these crashes. Artificial intelligence is used to analyze crash narratives for trip purposes such as traveling to grocery stores, bus stops, schools, or workplaces, offering contextual information not captured in standard crash reports. Specifically, this study will locally deploy an open-source large language model (e.g., Gemma or Grok) to serve as a traffic crash analysis agent, capable of addressing questions that help uncover the motivations behind pedestrian trips based on police crash narratives. By conducting all processing locally, this approach ensures the privacy of both pedestrians and drivers is preserved. Together, these methods distinguish between near-home and far-from-home crashes, highlight populations more frequently affected, and examine the role of broader urban development patterns. The findings will support city- and neighborhood-level safety strategies, helping target interventions on hazardous arterials and informing policies for improved safety.
- Record URL:
Language
- English
Project
- Status: Active
- Funding: $108,792.00
-
Contract Numbers:
69A3552348336
-
Sponsor Organizations:
Office of the Assistant Secretary for Research and Technology
University Transportation Centers Program
Department of Transportation
Washington, DC United States 20590 -
Managing Organizations:
Office of the Assistant Secretary for Research and Technology
Department of Transportation
1200 New Jersey Avenue, SE
Washington, DC United States 20590 -
Project Managers:
Stearns, Amy
-
Performing Organizations:
University of Tennessee, Knoxville
Center for Transportation Research (CTR)
Knoxville, TN United States 37996 -
Principal Investigators:
Parajuli, Saurav
Cherry, Chris
- Start Date: 20251201
- Expected Completion Date: 20261130
- Actual Completion Date: 0
- USDOT Program: University Transportation Centers
Subject/Index Terms
- TRT Terms: Artificial intelligence; Crash data; Data analysis; Pedestrian safety; Pedestrian vehicle crashes; Trip purpose
- Geographic Terms: Tennessee
- Subject Areas: Data and Information Technology; Pedestrians and Bicyclists; Safety and Human Factors;
Filing Info
- Accession Number: 01971453
- Record Type: Research project
- Source Agency: Center for Pedestrian and Bicyclist Safety
- Contract Numbers: 69A3552348336
- Files: UTC, RIP
- Created Date: Nov 17 2025 4:49PM